Fix memcpy32_sse2_unalign.
The whole point of mempcy32_sse2_unalign is that we didn't align dst128
and src128. So it's not safe at all to cast them back to dst and src.
That tells the compiler that dst/src are 128-bit aligned, and then it
autovectorizes the cleanup while-loop using that (false) knowledge with
aligned SSE instructions.
This leads to crashes on memcpy32_sse2_unalign_10, which is small enough
that we actually get non-16-byte aligned memory. The larger size
benches could be crashing too, but they're big enough allocations that
they're probably always 16-byte aligned anyway.
BUG=skia:2589
R=fmalita@chromium.org, mtklein@google.com
Author: mtklein@chromium.org
Review URL: https://codereview.chromium.org/291893008
git-svn-id: http://skia.googlecode.com/svn/trunk@14851 2bbb7eff-a529-9590-31e7-b0007b416f81
diff --git a/bench/MemcpyBench.cpp b/bench/MemcpyBench.cpp
index 0bb7b61..aec2a47 100644
--- a/bench/MemcpyBench.cpp
+++ b/bench/MemcpyBench.cpp
@@ -90,6 +90,8 @@
__m128i* dst128 = reinterpret_cast<__m128i*>(dst);
const __m128i* src128 = reinterpret_cast<const __m128i*>(src);
+ dst += 16 * (count / 16);
+ src += 16 * (count / 16);
while (count >= 16) {
__m128i a = _mm_loadu_si128(src128++);
__m128i b = _mm_loadu_si128(src128++);
@@ -103,9 +105,6 @@
count -= 16;
}
-
- dst = reinterpret_cast<uint32_t*>(dst128);
- src = reinterpret_cast<const uint32_t*>(src128);
}
while (count --> 0) {
@@ -122,6 +121,8 @@
static void memcpy32_sse2_unalign(uint32_t* dst, const uint32_t* src, int count) {
__m128i* dst128 = reinterpret_cast<__m128i*>(dst);
const __m128i* src128 = reinterpret_cast<const __m128i*>(src);
+ dst += 16 * (count / 16);
+ src += 16 * (count / 16);
while (count >= 16) {
__m128i a = _mm_loadu_si128(src128++);
__m128i b = _mm_loadu_si128(src128++);
@@ -136,14 +137,11 @@
count -= 16;
}
- dst = reinterpret_cast<uint32_t*>(dst128);
- src = reinterpret_cast<const uint32_t*>(src128);
while (count --> 0) {
*dst++ = *src++;
}
}
-// skia:2589: Crashing on ChromeOS Alex bot. TODO(mtklein): why?
-//BENCH(memcpy32_sse2_unalign, 10)
+BENCH(memcpy32_sse2_unalign, 10)
BENCH(memcpy32_sse2_unalign, 100)
BENCH(memcpy32_sse2_unalign, 1000)
BENCH(memcpy32_sse2_unalign, 10000)